Update TFLite concrete function conversion codes to have the associated (#2607)

* Update TFLite concrete function conversion codes to have the associated
trackable object.

Providing trackable objects is now recommended for the users. This makes the
concrete function conversion API be based on the new SavedModel importer, which
will enable new TensorFlow Lite features including variable support, resources
and variant tensor, and signature concept.

Author: abattery

tensorflow_addons/seq2seq/tests/decoder_test.py

@@ -113,7 +113,12 @@ def _decode(start_tokens, end_token):
     concrete_function = _decode.get_concrete_function(
         tf.TensorSpec([1], dtype=tf.int32), tf.TensorSpec([], dtype=tf.int32)
     )
-    converter = tf.lite.TFLiteConverter.from_concrete_functions([concrete_function])
+    if tf.__version__[:3] >= "2.7":
+        converter = tf.lite.TFLiteConverter.from_concrete_functions(
+            [concrete_function], _decode
+        )
+    else:
+        converter = tf.lite.TFLiteConverter.from_concrete_functions([concrete_function])
     converter.target_spec.supported_ops = [
         tf.lite.OpsSet.TFLITE_BUILTINS,
         tf.lite.OpsSet.SELECT_TF_OPS,